def main():
""" Reads command line arguments and starts either training, evaluation or inference. """
parser = argparse.ArgumentParser()
parser.add_argument(
'--config',
default='',
help='JSON-formatted file with configuration parameters')
parser.add_argument('--mode',
default='',
help='The operational mode - train|evaluate|predict')
parser.add_argument('--input',
default='',
help='The path to a PLY file to run through the model')
parser.add_argument('--device',
default='gpu',
help='The device to use - cpu|gpu')
parser.add_argument('--colors',
default='',
help='A file containing colors')
cla = parser.parse_args()
if cla.config == '':
print 'Error: --config flag is required. Enter the path to a configuration file.'
if cla.mode == 'evaluate':
inference.evaluate(cla.config, cla.device)
elif cla.mode == 'predict':
inference.predict(cla.config, cla.input, cla.device, cla.colors)
elif cla.mode == 'train':
training.train(cla.config)
else:
print 'Error: invalid operational mode - options: train, evaluate or predict'
def main():
flags = parse_flags()
hparams = parse_hparams(flags.hparams)
if flags.mode == 'train':
train.train(model_name=flags.model,
hparams=hparams,
class_map_path=flags.class_map_path,
train_csv_path=flags.train_csv_path,
train_clip_dir=flags.train_clip_dir,
train_dir=flags.train_dir)
elif flags.mode == 'eval':
evaluation.evaluate(model_name=flags.model,
hparams=hparams,
class_map_path=flags.class_map_path,
eval_csv_path=flags.eval_csv_path,
eval_clip_dir=flags.eval_clip_dir,
checkpoint_path=flags.checkpoint_path)
else:
assert flags.mode == 'inference'
inference.predict(model_name=flags.model,
hparams=hparams,
class_map_path=flags.class_map_path,
test_clip_dir=flags.test_clip_dir,
checkpoint_path=flags.checkpoint_path,
predictions_csv_path=flags.predictions_csv_path)
def testInference(self):
binary_path = os.path.join('../binaries', 'index-linux')
model_path = os.path.join('../test_data', 'model.json')
test_data_dir = os.path.join('../test_data')
tmp_dir = tempfile.mkdtemp()
inference.predict(binary_path, model_path, test_data_dir, tmp_dir)
with open(os.path.join(tmp_dir, 'data.json'), 'rt') as f:
ys_values = json.load(f)
# The output is a list of tensor data in the form of dict.
# Example output:
# [{"0":0.7567615509033203,"1":-0.18349379301071167,"2":0.7567615509033203,"3":-0.18349379301071167}]
ys_values = [list(y.values()) for y in ys_values]
with open(os.path.join(tmp_dir, 'shape.json'), 'rt') as f:
ys_shapes = json.load(f)
with open(os.path.join(tmp_dir, 'dtype.json'), 'rt') as f:
ys_dtypes = json.load(f)
self.assertAllClose(ys_values[0], [
0.7567615509033203, -0.18349379301071167, 0.7567615509033203,
-0.18349379301071167
])
self.assertAllEqual(ys_shapes[0], [2, 2])
self.assertEqual(ys_dtypes[0], 'float32')
# Cleanup tmp dir.
shutil.rmtree(tmp_dir)
def main(argv):
hparams = model.parse_hparams(flags.hparams)
if flags.mode == 'train':
def split_csv(scopes):
return scopes.split(',') if scopes else None
train.train(model_name=flags.model, hparams=hparams,
class_map_path=flags.class_map_path,
train_csv_path=flags.train_csv_path,
train_clip_dir=flags.train_clip_dir,
train_dir=flags.train_dir,
epoch_batches=flags.epoch_num_batches,
warmstart_checkpoint=flags.warmstart_checkpoint,
warmstart_include_scopes=split_csv(flags.warmstart_include_scopes),
warmstart_exclude_scopes=split_csv(flags.warmstart_exclude_scopes))
elif flags.mode == 'eval':
evaluation.evaluate(model_name=flags.model, hparams=hparams,
class_map_path=flags.class_map_path,
eval_csv_path=flags.eval_csv_path,
eval_clip_dir=flags.eval_clip_dir,
eval_dir=flags.eval_dir,
train_dir=flags.train_dir)
else:
assert flags.mode == 'inference'
inference.predict(model_name=flags.model, hparams=hparams,
class_map_path=flags.class_map_path,
inference_clip_dir=flags.inference_clip_dir,
inference_checkpoint=flags.inference_checkpoint,
predictions_csv_path=flags.predictions_csv_path)
def main():
flags = parse_flags()
hparams = parse_hparams(flags.hparams)
if flags.mode == 'train':
utils.resample(sample_rate=flags.sample_rate, dir=flags.train_clip_dir, csv_path=flags.train_csv_path)
train.train(model_name=flags.model, hparams=hparams,
class_map_path=flags.class_map_path,
train_csv_path=flags.train_csv_path,
train_clip_dir=flags.train_clip_dir+'/resampled',
train_dir=flags.train_dir, sample_rate=flags.sample_rate)
elif flags.mode == 'eval':
#TODO uncomment
#utils.resample(sample_rate=flags.sample_rate, dir=flags.eval_clip_dir, csv_path=flags.eval_csv_path)
evaluation.evaluate(model_name=flags.model, hparams=hparams,
class_map_path=flags.class_map_path,
eval_csv_path=flags.eval_csv_path,
eval_clip_dir=flags.eval_clip_dir+'/resampled',
checkpoint_path=flags.checkpoint_path)
else:
assert flags.mode == 'inference'
utils.resample(sample_rate=flags.sample_rate, dir=flags.test_clip_dir, csv_path='test')
inference.predict(model_name=flags.model, hparams=hparams,
class_map_path=flags.class_map_path,
test_clip_dir=flags.test_clip_dir,
checkpoint_path=flags.checkpoint_path,
predictions_csv_path=flags.predictions_csv_path)
def executor(locals_dict):
# Updating all user parameters
start = timer()
global a
a = 0
def update_config(current_path, config_main):
_dict = {}
global a
for key, value in config_main.items():
if key in locals_dict:
_dict[key] = locals_dict[key]
elif (current_path + "__" + key) in locals_dict:
_dict[key] = locals_dict[(current_path + "__" + key)]
print(current_path + "__" + key)
elif isinstance(value, dict):
if a == 1:
if len(current_path) > 0:
_dict[key] = update_config(current_path + "__" + key,
value)
else:
_dict[key] = update_config(current_path + key, value)
else:
a = 1
_dict[key] = update_config(current_path, value)
a = 0
else:
_dict[key] = config_main[key]
return _dict
current_config = update_config("", config_main)
with open(f'./assets/config-{now}.yaml', 'w') as file:
yaml.safe_dump(current_config, file)
with open('config.yaml', 'w') as config_data:
yaml.safe_dump(current_config, config_data)
if current_config['limit_gpu'] is not False:
limit_gpu()
if current_config['mode'] == 'training':
print("[INFO]...Starting Training Job")
train(current_config)
if current_config['mode'] == 'inference':
print("[INFO]...Starting Inference Job")
predict(current_config)
end = timer()
print(
f"[INFO]...Time taken by {current_config['mode']} is {(end - start)/60:.2f} min(s)"
)
def run():
"""Runs dilated residual network model in either train or predict mode"""
config = parse_args()
if not os.path.isdir(config.logs):
os.makedirs(config.logs)
if config.mode == 'train':
train(config)
elif config.mode == 'predict':
predict(config)
else:
ValueError("Mode must be either train or predict")
def testInferenceWithNonExistingOutputNameFile(self):
binary_path = os.path.join('../binaries', 'tfjs-inference-linux')
model_path = os.path.join('../test_data', 'model.json')
test_data_dir = os.path.join('../test_data')
tmp_dir = tempfile.mkdtemp()
inference.predict(binary_path, model_path, test_data_dir, tmp_dir, tf_output_name_file='non_exist.json')
with self.assertRaises(FileNotFoundError):
with open(os.path.join(tmp_dir, 'data.json'), 'rt') as f:
json.load(f)
# Cleanup tmp dir.
shutil.rmtree(tmp_dir)
def do_infer(model, data, device):
if _g.args.quiet:
return
print("Infering {} test examples...".format(_g.args.examples))
for i in range(_g.args.examples):
print("Case {}:".format(i+1))
index = random.randrange(0, len(data.examples))
sample = data.examples[index]
if _g.args.debug:
_i.predict_and_plot(model, sample.before, sample.after, device)
else:
_i.predict(model, sample.before, sample.after, device)
def fileHandler():
if flask.request.method == 'GET':
return jsonify({
"reponse":
"Please POST a .txt file containing submission urls at this endpoint"
})
if flask.request.method == 'POST':
file = request.files['file']
# print(file)
urls = file.read().splitlines()
correct = 0
top3_correct = 0
total = len(urls)
for url in urls:
text, flair = get_data(url.decode())
pred, confi = predict(text)
## Something fishy ewww
if flair in pred or pred not in flairs:
top3_correct += 1
if flair == pred[0]:
correct += 1
# print(correct)
# print(total)
return jsonify({
"top-1 accuracy": '%.4f' % (correct / total),
"top-3 accuracy": '%.4f' % (top3_correct / total)
})
def preprocess():
WIDTH = 484
HEIGHT = 240
ENSEMBLE_N = 3
# GET COLOR ENCODING AND ITS INDEX MAPPING
colors = loadmat('../data/color150.mat')['colors']
root = '..'
names = {}
with open('../data/object150_info.csv') as f:
reader = csv.reader(f)
next(reader)
for row in reader:
names[int(row[0])] = row[5].split(";")[0]
idx_map = create_idx_group()
colors, names = edit_colors_names_group(colors, names)
# SETUP MODEL
cfg_path = os.path.join('..', 'config',
'ade20k-mobilenetv2dilated-c1_deepsup.yaml')
#cfg_path="config/ade20k-resnet18dilated-ppm_deepsup.yaml"
model = setup_model(cfg_path, root, gpu=0)
model.eval()
# GET DATA AND PROCESS IMAGE
data = np.load(os.path.join('..', 'test_set', 'cls1_rgb.npy'))
data = data[:, :, ::-1]
img = ImageLoad_cv2(data, WIDTH, HEIGHT, ENSEMBLE_N, True)
# MODEL FEED
predictions = predict(model, img, ENSEMBLE_N, gpu=0, is_silent=False)
return predictions, colors, names, idx_map
def reactRequestHandler():
if flask.request.method == 'GET':
return jsonify(
{"reponse": "please POST a valid submission url at this endpoint"})
if flask.request.method == 'POST':
data = request.get_json()
url = data['url']
# print(url) # for debuggin
try:
text, flair = get_data(url)
pred, confi = predict(text)
result = [
str(cls) + " : " + str('%.4f' % confidence)
for cls, confidence in zip(pred, confi)
]
return jsonify({
"top_3_pred": " ".join(result),
"flair": flair,
"text": text
})
except:
return jsonify({
"response":
"Oops something went wrong! Check your internet connection and/or url"
})
def infer_fitness_and_frequencies(obs_reads_df, proposed_gts, out_dir, options = ''):
'''
Uses obs_reads_df, which describes observed frequencies of mutations (symbols) in distinct linkage groups, and proposed_gts, a list of full-length genotypes that may occur in the population.
Infers the frequency of each full-length genotype at each timepoint, and a fitness value for each full-length genotype. Saves output objects to out_dir, and returns them in a dict.
'''
return inf.predict(obs_reads_df, proposed_gts, out_dir, options = options)
def test_wiki2vec(self):
acc = 0
n = 0
dataloader = DataLoader(dataset=self.dataset,
batch_sampler=self.batch_sampler,
collate_fn=collate_wiki2vec)
for batch_num, batch in enumerate(dataloader):
batch = u.tensors_to_device(batch, self.device)
labels_for_batch = self._get_labels_for_batch(batch['label'],
batch['candidate_ids'])
predictions = predict(embedding=self.embedding,
token_idx_lookup=self.token_idx_lookup,
p_prior=0 if self.use_adaptive_softmax else batch['p_prior'],
model=self.model,
batch=batch,
ablation=self.ablation,
entity_embeds=self.model.entity_embeds,
use_wiki2vec=True)
acc += int((labels_for_batch == predictions).sum())
batch_size = len(predictions)
n += batch_size
self.experiment.record_metrics({'accuracy': acc / n,
'TP': acc,
'num_samples': n})
return acc, n
def main_pipeline(port):
"""
Runs main pipeline: motion -> take picture -> classify -> action if the input on pin_motion is raising high.
:param port: pin of the motion sensor. In this case pin_motion. format: int
:return:
"""
global pap_flag
logger("INFO: motion detected")
if pap_flag:
logger("WARNING: main_pipeline is already running")
return
pap_flag = 1
img = take_pic()
predicted_class = inference.predict(img, model)
if predicted_class == 2:
logger("INFO: pigeon detected")
rotate_motor(rotation_time)
if predicted_class == 0:
logger("INFO: human detected")
time.sleep(180)
if predicted_class == 1:
logger("INFO: no one detected")
inference.save_image(img, predicted_class)
pap_flag = 0
return
def send_prediction(im):
bands = len(im.getbands())
imdata = np.array(im.getdata()).reshape(im.size[0], im.size[1], bands)
prediction = predict(net, colors, imdata)
strio = StringIO.StringIO()
prediction.save(strio, 'PNG')
strio.seek(0)
return send_file(strio, mimetype='image/png')
def send_prediction(im):
bands = len(im.getbands())
imdata = np.array(im.getdata()).reshape(im.size[0], im.size[1], bands)
prediction = predict(net, colors, imdata)
strio = StringIO.StringIO()
prediction.save(strio, 'PNG')
strio.seek(0)
return send_file(strio, mimetype='image/png')
def predict_process(path):
flist = get_filelist(path)
flist.sort()
for ts_file in flist:
if not is_infile(csv_file, ts_file):
with open(csv_file, mode='a') as csv_file_predict:
csv_write = csv.writer(csv_file_predict, delimiter=',')
csv_write.writerow(
[ts_file, predict(os.path.join(path, ts_file))])
def pred():
im = Image.open(request.files['image'])
bands = len(im.getbands())
imdata = np.array(im.getdata()).reshape(im.size[0], im.size[1], bands)
prediction = predict(net, colors, imdata)
strio = StringIO.StringIO()
prediction.save(strio, 'PNG')
strio.seek(0)
return send_file(strio, mimetype='image/png')
def do_real_input(model, device):
if _g.args.quiet:
return
for i in range(_g.args.examples):
print("Case {}:".format(i+1))
before = list(input('Input text: '))
after = input('Target? (y/n): ')
while after != "y" and after != "n":
after = input('Target? (y/n): ')
if after == "y":
after = list(input('Target text: '))
if _g.args.debug:
_i.predict_and_plot(model, before, after, device)
else:
_i.predict(model, before, after, device)
else:
_i.predict(model, before, device=device)
def __get_pred_image(self):
_, hint = self.painter.get_image()
self.__status_update(1)
line = self.origin_line
hint = Image.fromarray(hint)
img = predict(line, hint, None)
if not isinstance(img, Image.Image):
popup_err_dialog('Inference Err', img)
return img
def handle_tests(test: Test):
if test.ground_truth[0] != 5:
with open(f"/user_tests/{test.id}{datetime.datetime.now()}_gt.json",
'w') as f:
json.dump(test.json(), f)
return {"message": test.ground_truth[0]}
else:
pred = predict(test.text)
test.result = pred[1]
test.softmax = pred[2]
return test
def predict_upload():
#read the image from static directory
image = cv2.imread('static/uploads/prediction_image.jpg')
#get the predictions
stage1, stage2, stage3, final = predict(image)
#write the output to the static folder
cv2.imwrite('static/result/stage1.jpg', stage1)
cv2.imwrite('static/result/stage2.jpg', stage2)
cv2.imwrite('static/result/stage3.jpg', stage3)
cv2.imwrite('static/result/final.jpg', final)
return render_template('predict.html')
def home():
form = InputForm()
prediction_message = Prediction_Message()
if form.validate_on_submit():
prediction_message.initial, prediction_message.pred1, prediction_message.pred2, prediction_message.pred3 = \
predict(model, form.tweet.data, glove.get_glove_emb(), sequence_length=62, use_gpu=False)
# if form.tweet.data == 'helloworld':
# flash('Entered correct input!', 'success')
# return redirect(url_for('home'))
# else:
# flash('Invalid Tweet!', 'danger')
return render_template('home.html', title='DeepFeels', form = form, prediction_message = prediction_message)
def show_image(image_name):
print("hahahah", image_name)
fig = Figure()
image = Image.open(UPLOAD_FOLDER + '/' + image_name)
model = get_model()
top_probability, top_class = predict(image, model, topk=5)
with open('categories.json') as f:
cat_to_name = json.load(f)
fig = view_classify(image, top_probability, top_class, cat_to_name)
output = io.BytesIO()
FigureCanvasAgg(fig).print_png(output)
return Response(output.getvalue(), mimetype="image/png")
def hello_world():
if request.method == 'GET':
return render_template('index.html', value="with food")
if request.method == 'POST':
if 'file' not in request.files:
print("file not uploaded")
return
file = request.files['file']
image = file.read()
bboxes, labels, scores = predict(image_bytes=image)
return render_template('result.html',
labels=labels,
scores=scores,
boxes=bboxes)
def main():
flags = parse_flags()
hparams = parse_hparams(flags.hparams)
if flags.mode == 'train':
utils.resample(sample_rate=flags.sample_rate,
dir=flags.train_clip_dir,
csv_path=flags.train_csv_path)
train.train(model_name=flags.model,
hparams=hparams,
class_map_path=flags.class_map_path,
train_csv_path=flags.train_csv_path,
train_clip_dir=flags.train_clip_dir + '/resampled',
train_dir=flags.train_dir,
sample_rate=flags.sample_rate)
elif flags.mode == 'eval':
#TODO uncomment
#utils.resample(sample_rate=flags.sample_rate, dir=flags.eval_clip_dir, csv_path=flags.eval_csv_path)
evaluation.evaluate(model_name=flags.model,
hparams=hparams,
class_map_path=flags.class_map_path,
eval_csv_path=flags.eval_csv_path,
eval_clip_dir=flags.eval_clip_dir + '/resampled',
checkpoint_path=flags.checkpoint_path)
else:
assert flags.mode == 'inference'
utils.resample(sample_rate=flags.sample_rate,
dir=flags.test_clip_dir,
csv_path='test')
inference.predict(model_name=flags.model,
hparams=hparams,
class_map_path=flags.class_map_path,
test_clip_dir=flags.test_clip_dir,
checkpoint_path=flags.checkpoint_path,
predictions_csv_path=flags.predictions_csv_path)
def line_segment(image, img, output_path):
"""
img: image needed to segment to lines.
output_path: output folder of segmented images.
:return: output image pathes
"""
#cv2.imshow("Original", img)
#cv2.waitKey()
# (1) Segment Lines
line_segmentation = LineSegmentation(img=img, output_path=output_path)
lines = line_segmentation.segment()
# (2) Save lines to file
output_image_path = line_segmentation.save_lines_to_file(lines)
percTot = 0
count = 0
text = ""
for m in lines:
pd, pbperc = predict(cv2.cvtColor(m, cv2.COLOR_GRAY2RGB))
#cv2.imshow("Line",m)
#cv2.waitKey()
percTot += pbperc
count += 1
text += pd + '\n'
if count != 0:
perc = percTot / count
if count == 0:
pd, pbperc = predict(
cv2.cvtColor(cv2.imread(image), cv2.COLOR_GRAY2RGB))
perc = pbperc
text = pd
head, tail = ntpath.split(image)
image = tail or ntpath.basename(head)
filename = os.path.splitext(image)[0]
fout = open('output/' + filename + '_output.txt', 'w')
fperc = open('output/' + filename + '_probability.txt', 'w')
fout.write(text)
fperc.write(f"{perc:.2f}")
def single_mode():
'### Single Text & Sentiment'
'#### Text here:'
input_text = st.text_area('', max_chars=144)
'#### Sentiment:'
sentiment = st.selectbox('', ('negative', 'neutral', 'positive'))
if input_text:
d = {'text': input_text, 'sentiment': sentiment}
df = pd.DataFrame(data=d, index=[0])
pred = inference.predict(df, model=model)
'### Words explain sentiment:'
st.write(pred[0])
def vision():
if not request.json or not 'url' in request.json:
abort(400)
# get url
url = request.json['url']
t0 = time.time()
# vectorize image
x = vectorize(url)
# predict
with graph.as_default():
r = predict(x, model, ['cat', 'dog'])
t1 = time.time()
r['elapsed'] = t1 - t0
return jsonify(r), 201
def send():
if request.method == 'POST':
img_file = request.files['img_file']
if img_file and allowed_file(img_file.filename):
filename = secure_filename(img_file.filename)
img_file.save(os.path.join(app.config['UPLOAD_FOLDER'], filename))
img_url = UPLOAD_FOLDER + filename
# inferenceを呼び出し、Captionを生成する
img_caption = inference.predict(img_url)
return render_template('index.html',
img_url=img_url,
img_caption=img_caption)
else:
return ''' <p>許可されていない拡張子です</p> '''
else:
return redirect(url_for('index'))
def index():
if request.method == 'POST':
text = request.form['text']
# remove static directory if it exists
for files in glob.glob(os.path.join("static", "*")):
os.remove(files)
name, img, path = predict(text)
img.save(os.path.join("static", "molecule.png"))
path.save(os.path.join("static", "pathway.png"))
result = {
'mol_name': name,
'mol_path': os.path.join("static", "molecule.png"),
'pathway_path': os.path.join("static", "pathway.png")
}
return render_template('show.html', result=result)
return render_template(
'index.html') # by default it will look in templates folder
def make_prediction(net, colors, im, threshold, outfile):
bands = len(im.getbands())
imdata = np.array(im.getdata()).reshape(im.size[0], im.size[1], bands)
predicted = predict(net, colors, threshold, imdata)
predicted.save(outfile, 'PNG')
caffe.set_mode_gpu()
if args.gpu is not None:
caffe.set_device(args.gpu)
net = caffe.Net(args.model,
args.weights,
caffe.TEST)
pixels_correct = 0
pixels_complete = 0
pixels_predicted_fg = 0
pixels_actual_fg = 0
outputs = []
for i in range(0, len(test_data)):
prediction_image = predict(net, colors)
image = net.blobs['data'].data
image = np.squeeze(image[0, :, :, :])
label = net.blobs['label'].data
label = np.squeeze(label).astype(np.uint8)
predicted = net.blobs['prob'].data
predicted = np.squeeze(predicted[0, :, :, :])
metrics = complete_and_correct(predicted, label, 3, 0.5)
result = {'index': i, 'metrics': metrics}
print(result)
outputs.append(result)
pixels_correct += sum(metrics['pixels_correct'])
pixels_complete += sum(metrics['pixels_complete'])
